You are on page 1of 12

International Journal

of Behavioral Development
http://jbd.sagepub.com/

Multivariate latent change modeling of developmental decline in academic intrinsic math motivation
and achievement: Childhood through adolescence
Adele Eskeles Gottfried, George A. Marcoulides, Allen W. Gottfried, Pamella H. Oliver and Diana Wright Guerin
International Journal of Behavioral Development 2007 31: 317
DOI: 10.1177/0165025407077752
The online version of this article can be found at:
http://jbd.sagepub.com/content/31/4/317

Published by:
http://www.sagepublications.com

On behalf of:

International Society for the Study of Behavioral Development

Additional services and information for International Journal of Behavioral Development can be found at:
Email Alerts: http://jbd.sagepub.com/cgi/alerts
Subscriptions: http://jbd.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://jbd.sagepub.com/content/31/4/317.refs.html

>> Version of Record - Jul 2, 2007


What is This?

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

International Journal of Behavioral Development


2007, 31 (4), 317327

2007 The International Society for the


Study of Behavioural Development

http://www.sagepublications.com

DOI: 10.1177/0165025407077752

Multivariate latent change modeling of developmental decline in


academic intrinsic math motivation and achievement: Childhood
through adolescence
Adele Eskeles Gottfried
California State University, Northridge, USA

George A. Marcoulides,
Allen W. Gottfried, Pamella H. Oliver,
and Diana Wright Guerin
California State University, Fullerton, USA

Research has established that academic intrinsic motivation, enjoyment of school learning without
receipt of external rewards, significantly declines across childhood through adolescence. Math intrinsic motivation evidences the most severe decline compared with other subject areas. This study
addresses this developmental decline in math intrinsic motivation, and also serves as a resource for
applied researchers by providing exemplary illustrations of approaches to longitudinal modeling.
Using a multivariate latent change model, the longitudinal relationship between academic intrinsic
math motivation and math achievement among participants (n = 114) aged 917 years was examined
to explain this motivational decline. On average, both math motivation and achievement decreased
over time. This study reveals that math achievement is a significant contributor to the developmental
decline in intrinsic math motivation from childhood through adolescence. In addition, academic
intrinsic math motivation was found to be related to initial and later levels of mathematics achievement. These findings enhance understanding of developmental processes whereby early motivation
and achievement are related to subsequent declines in mathematics.
Keywords: academic intrinsic motivation; longitudinal modeling; math motivation

Motivation is a critically important factor for academic


learning and achievement across childhood through adolescence (Elliot & Dweck, 2005). Academic intrinsic motivation
plays a particularly important role with regard to school
learning and achievement because of its inherent relatedness
to cognitive processing and mastery (Berlyne, 1971; A.E.
Gottfried, 1985; Hunt, 1971; White, 1959) as well as relationships to academic competency. Academic intrinsic motivation
is defined as enjoyment of school learning characterized by an
orientation toward mastery, curiosity, persistence, taskendogeny, and the learning of challenging, difficult, and novel
tasks (A.E. Gottfried, 1985). Further, from childhood through
adolescence, across varied populations, those with higher
academic intrinsic motivation have been found to be more
competent in school, generally evidencing significantly greater
academic achievement, more positive perceptions of their
academic competency, lower academic anxiety, and less extrinsic motivation (A.E. Gottfried, 1985, 1990; A.E. Gottfried,
Fleming, & Gottfried, 1994; A.E. Gottfried & Gottfried, 2004;
A.W. Gottfried, Gottfried, Bathurst, & Guerin, 1994; A.W.
Gottfried, Gottfried, Cook, & Morris, 2005). Nonetheless,
regardless of the well-documented findings concerning the
importance of motivation for learning and achievement, there

is a ubiquitous and alarming finding across childhood and


adolescence, that being a significant developmental decline in
academic motivation, including academic intrinsic motivation.
The purposes of this study were to: (1) investigate the
prediction that math achievement is a significant contributor
to the developmental decline in intrinsic math motivation
within a longitudinal study encompassing data from childhood
through late adolescence; and (2) address this issue through a
multivariate longitudinal model. We also expect this study to
serve as a resource for applied researchers interested in utilizing such multivariate longitudinal methods in their own
research by providing exemplary illustrations of good practice
approaches to longitudinal modeling. Of considerable concern
for the interpretation of longitudinal models such as those
examined here are the consequences of how time is coded in
order to facilitate the judgment of model parameters. Although
the potential problems of various coding choices have been
pointed out before (Biesanz, Deeb-Sossa, Papadakis, Bollen,
& Curran, 2004; Raykov & Marcoulides, 2006; Rogosa &
Willett, 1985) it appears that the issue is not clearly understood and often eludes researchers. Indeed, although the
problem of how to code time might appear somewhat arbitrary
and trivial, it turns out that the interpretation of the resulting

Correspondence should be sent to Professor A.E. Gottfried, Dept of


Educational Psychology, California State University, Northridge,
CA 91330, USA; e-mail: adele.gottfried@csun.edu; or Dr G.A.
Marcoulides, Dept of Information Systems and Decision Sciences,
California State University, Fullerton, CA 92634, USA; e-mail:
gmarcoulides@fullerton.edu

This research was partially supported by Faculty Research Grants


from California State University, Northridge and Fullerton, the
Spencer Foundation, and the Thrasher Foundation. Gratitude is
extended to the participants in the Fullerton Longitudinal Study and
their families for their involvement.

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

318

GOTTFRIED ET AL. / LATENT CHANGE ANALYSIS OF DECLINE IN MOTIVATION

solution certainly is not. Ultimately, as will become clear, our


recommendations concerning the choice of appropriate
modeling approach and coding of time will be based on the
ease with which model parameter estimates can be readily
interpreted, always keeping of course in mind the substantive
questions of the research study.
Math was selected as the academic domain studied for
several reasons. Previous research on academic intrinsic
motivation from childhood and adolescence has revealed a
significant decline for the subject areas of reading, math,
and science, as well as for school in general. However, the
subject area with the greatest and steepest decline was math
(A.E. Gottfried, Fleming, & Gottfried, 2001). Hence, this
subject area appears to be the most at-risk for developmental decline. A decline in math-related motivation has
also been found by other researchers (Wigfield, Eccles,
Schiefele, Roeser, & Davis-Kean, 2006). For example,
Fredericks and Eccles (2002) documented declining math
interest across the transition to junior high school, and
Eccles and Midgley (1989) and Jacobs, Lanza, Osgood,
Eccles, and Wigfield (2002) found that students had declining perception of math competence as they advanced across
grade levels. In a review of the literature, Middleton and
Spanias (1999) concluded that there was ample evidence of
a developmental decline in enjoyment of math from elementary through high school.
In addition to the developmental decline in math-related
motivation occurring pervasively across ages, populations,
and studies, there is cause for concern at national and international levels as well. International comparative assessments,
such as those obtained in the Third International Mathematics and Science Study (TIMSS), have found that American
children are far behind those of other countries (Tatsuoka,
Corter, & Tatsuoka, 2004). In this research, the USA places
in the middle of a field of 40 countries. Increasing the math
expertise of children growing up in the USA is recognized as
an area of special national need. In a recently published
volume, it was explicitly stated that there is a critical need for
math skills in our increasingly technological society (National
Academy of Sciences, 2005). By implication, then, the USA
will continue to fall behind other nations in developing the
mathematical expertise of students needed for contemporary
professions, sciences, and the economy. Hence, the mandate
is clear if we are to prevent further deterioration regarding
math motivation and encourage entry into math-related
professions.
Besides developmental declines in math motivation,
developmental declines have also been observed in general
aspects of academic motivation. These include intrinsic motivation not distinguished by subject area (Harter, 1981; Lepper,
Corpus, & Iyengar, 2005; Lepper & Henderlong, 2000;
Spinath & Spinath, 2005); perception of competence and selfconcept (Anderman & Maehr, 1994; Eccles & Midgley, 1989,
1990; Fredericks & Eccles, 2002), and changes in perceived
value of achievement and goal orientations (Wigfield &
Wagner, 2005). Overall, the phenomenon of declining
academically related motivation is pervasive.
Prior explanations of this decline have included a lack of fit
between school environments and the developmental needs of
students (Eccles & Midgley, 1989, 1990), and an increase in
the controlling and extrinsic atmosphere of classrooms and
schools encountered by students as they transition from
elementary to middle school (Eccles, Wigfield, & Schiefele,

1998; Harter, 1981; Lepper & Henderlong, 2000; Lepper,


Sethi, Dialdin, & Drake, 1997). However, these explanations
are based on the nonrandom entry of students to new schools.
Hence, if students enter new schools with greater or lesser
controlling atmospheres, one cannot separate the school
environment from pre-existing selection factors. Other general
explanations have also included an increase in social comparison between students as they develop (Wigfield & Wagner,
2005; Wigfield et al., 2006), an increasing emphasis on
performance as compared with mastery goals across the grades
(Anderman & Maehr, 1994; Eccles & Midgley, 1989; Lepper
& Henderlong, 2000; Wigfield & Wagner, 2005), and an
absence of optimal challenges and lack of the relevance of
curriculum (Lepper & Henderlong, 2000). Regardless of these
possibilities, none has been oriented specifically to math motivation itself. In their review article, Middleton and Spanias
(1999) concluded that what is lacking in research is studying
the development and change of math motivation over time.
This is the specific issue examined in the present research.
It has recently been noted that research on developmental
declines in competency beliefs is normative, i.e., reporting
mean level changes (Wigfield et al., 2006). Therefore, the
research on age-related developmental declines in motivation
lacks a developmental approach regarding explanations of
this adverse shift. This study goes beyond the reporting of
mean differences by examining developmental processes
accounting for this decline. We examine the role of childrens
developmental patterns of math achievement as contributing
to the developmental decline in math motivation. Thus, we
examine the co-occurrence of age-related developmental
patterns in two domains: math motivation and math
achievement.
Prior explanations have not directly examined the role of
academic achievement as a contributor to the developmental
decline in math motivation. There is both a theoretical and
empirical foundation for examining the associations between
math motivation and achievement. Theoretically, intrinsic
motivation is considered to be an integral aspect of learning
and cognition inasmuch as those who inherently enjoy learning
engage in activities that further advance their cognition and
mastery (Berlyne, 1971; A.E. Gottfried, 1985; Hunt, 1971;
White, 1959). Empirically, there is an enormous literature
documenting significant, positive relationships between motivation and achievement, intrinsic math motivation and math
achievement, and between math attitudes and achievement
both within populations in the USA and internationally
(A.E. Gottfried, 1985, 1990; A.E. Gottfried & Gottfried,
2004; A.E. Gottfried et al., 1994; ODwyer, 2005; Shen, 2002;
Uguroglu & Walberg, 1979). By examining achievement as a
contributor to the decline in academic intrinsic motivation, a
direct index of competence is available, rather than the indirect
indices typically used in developmental research, such as
perception of competence.
In the present longitudinal investigation, known as the
Fullerton Longitudinal Study (FLS), both math motivation
and achievement have been assessed over a long-term period
from childhood through late adolescence. This allowed us to
examine the contribution of changes in math achievement to
changes in math motivation. Moreover, in the FLS, not only
is there an apparent decline in intrinsic math motivation, but
there is also a decline in math achievement across childhood
through adolescence. This dual decline provided a compelling rationale to examine the contribution of achievement

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2007, 31 (4), 317327

to the decline in intrinsic math motivation. It should be


noted that the developmental decline in math achievement is
not unique to the FLS, but is consistent with international
trends found in the TIMSS data. For example, Kelly (2002)
reported that fewer children in the USA reached specific
benchmarks (10th, 25th, 50th, and 75th percentiles) in the
eighth compared with fourth grade, indicating a decline in
math achievement with increasing grade level. Further, math
course grades have been shown to decrease across grade
levels from elementary through junior high school (Eccles &
Midgley, 1989); and growth rate in math achievement has
been found to decline from grades 3 through 7 (Ding &
Davison, 2004).

Method
Participants
The FLS furnished the database for the present study. The
FLS is a contemporary investigation that was initiated in 1979
with 130 infants and their families. All children who entered
the study had been term babies of normal birth weight and had
no neurological or visual abnormalities. During the course of
the study, children were assessed in the university laboratory
at 6-month intervals from 1 to 3.5 years and at yearly intervals
beginning at age 5 to 17 years. At each assessment a comprehensive battery of standardized measures was administered to
examine development across a broad variety of domains. The
retention rate of this sample was substantial with no less than
80% of the original sample returning at any assessment. There
was no evidence of attrition bias throughout the course of the
study (Guerin & Gottfried, 1994; Guerin, Gottfried, Oliver, &
Thomas, 2003). At the outset of the investigation, the participants resided typically within an hour of the research site.
Because geographic mobility is common in the course of
development and family life, the study population eventually
resided throughout the USA and even abroad (A.W. Gottfried,
Gottfried, & Guerin, 2006). This is important to note because
the developmental trends obtained in the FLS are not
confounded with a specific school or school district, teacher,
or curriculum. The developmental trajectories to be reported
below generalize across schools, curricula, teachers, and
geographic area.
The data analyzed here comprise academic achievement
(assessed from ages 9 to 17 years) and academic intrinsic
motivation (assessed from ages 9 to 17 years see measures
section for further details). The current analyses are based on
data from 114 participants. The socioeconomic status of the
sample represented a wide, middle-class range, from semiskilled workers through professionals, as determined by the
Hollingshead Four-Factor Index of
Social Status
(Hollingshead, 1975; also see A.W. Gottfried, 1985;
A.W. Gottfried, Gottfried, Bathurst, Guerin, & Parramore,
2003). The mean Hollingshead Social Status Index was 45.6
(SD = 11.9) at the initiation of the FLS and 48.6 (SD = 11.4)
at the 17-year assessment. At the initiation of the study, participants were predominantly European American (90%) and also
from other ethnic backgrounds. The percentages of males and
females were 52 and 48, respectively. For further details
concerning sample characteristics and study design, see A.W.
Gottfried and Gottfried (1984), A.W. Gottfried et al. (1994,
2006), and Guerin, Gottfried, Oliver, & Thomas (2003).

319

Measures
Academic intrinsic motivation. Academic intrinsic motivation,
as defined above, was assessed with the Childrens Academic
Intrinsic Motivation Inventory (CAIMI; A.E. Gottfried, 1986)
a psychometrically established scale that provides four subject
area subscales (reading, math, social studies, and science) as
well as a subscale for school in general (A.E. Gottfried, 1986).
Mathematics was selected as the subject area studied based on
previous research indicating that it is a unique subject area
showing specific relations to math achievement and other math
criteria (A.E. Gottfried, 1985, 1990), as well as past research
that math evidences the sharpest decline across childhood
through adolescence (A.E. Gottfried et al., 2001). The CAIMI
was administered in the FLS to each participant individually
in the laboratory assessment at ages 9, 10, 13, 16, and 17 years.
Academic achievement. Math achievement was assessed yearly
from ages 9 to 17 using the individually administered
WoodcockJohnson Psycho-Educational Battery from ages 9
through 10 years (Woodcock & Johnson, 1977) and the revised
WoodcockJohnson Psycho-Educational Battery (Woodcock
& Johnson, 1989) from ages 11 to 17 years. This instrument
was selected as it provided consistency in the measurement of
achievement across childhood through adolescence. Analyses
were conducted on the Broad Math grade percentile score
which comprises two subtests (calculation and applied
problems). The advantage of the grade percentile is that it
furnishes a score correcting for grade level at a given age
(Woodcock & Johnson, 1977, 1989). Scores for the longitudinal data of the single cohort are based on the established
WoodcockJohnson cross-sectional norms. Thus, observed
change in achievement is the relative change in standing and
not absolute value in change over time. This is in accord with
the aforementioned literature.

Modeling developmental trends using structural


equation modeling
Modeling change within a structural equation modeling
(SEM) framework is a relatively recent and popular approach
for studying developmental trends. This is because SEM offers
a greater degree of flexibility in testing a variety of hypotheses
concerning the developmental trends than a number of other
more traditional techniques like repeated measures analysis of
variance (Raykov & Marcoulides, 2006). In this article we refer
to this specific SEM framework as latent change analysis
(LCA), because its focus is the study of change over time. In
simple terms, a LCA model essentially resembles a basic
confirmatory factor analysis model in which the latent variables are interpreted as chronometric variables representing
individual differences over time. It is important to note that
LCA models have also been referred to in the literature under
various alternative names, such as latent growth curve
models, latent curve analysis models, or just growth curve
models. We prefer the reference LCA models, to emphasize
that the models are applicable to all cases in which one is
interested in studying change (either growth or decline), even
those with a more complex pattern of change such as growth
followed by decline or vice versa (for further details see also
Duncan, Duncan, Strycker, Li, & Alport, 1999; Raykov &
Marcoulides, 2006).
Two specific kinds of longitudinal modeling strategies can

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

320

GOTTFRIED ET AL. / LATENT CHANGE ANALYSIS OF DECLINE IN MOTIVATION

be utilized to address the issues considered in this study, the


so-called Level-and-Shape (LS) model and the Intercept-andSlope (IS) model. The LS model was described by McArdle
(1988) but does not seem to have attracted as much widespread attention as the currently popular IS model, despite the
fact that it has a number of advantages over the IS model
(Raykov & Marcoulides, 2006). A main assumption of the IS
model is that change occurs in a specific fashion (e.g., in either
a linear, quadratic, or cubic fashion). Unfortunately, the actual
change process may be quite difficult to precisely model utilizing a specific trajectory. Because of such concerns, we believe
that the less restrictive LS model is preferable because it would
be expected to fit the data better (mainly because it does not
impose a specific type of trajectory upon the pattern of change
occurring in the study). In addition, we note that the IS model
is obtained as a special case of the LS model when the coding
of time is fixed according to the time at which the repeated
measurements were obtained (discussed in detail below). For
illustrative purposes, in this article we fit both the LS and the
IS model and provide insight concerning some difficulties that
might arise when modeling developmental data.
Assuming that the series of five repeated ordered waves of
measurements on intrinsic math motivation (using the
CAIMI) are represented as Yit (where the index i represents
each observed individual in the study and t represents the timeordered measurements on the CAIMI), a simple longitudinal
model equation for describing an individuals development
over the repeated measurements (also called a level 1 or withinperson model) can be written as:
Yit = yi + yit + it

M9

M10

yi = y + yi

(2)

yi = y + yi

(3)

It is usually assumed that there is no covariance structure


between the residuals of the longitudinally observed variables,
implying that the covariance matrix is diagonal. The level 2
model is also often called an unconditional model if no other
predictors are believed to account for the variation in the individual parameters of the trajectories. Such an unconditional
LCA model for the five repeated measurements on intrinsic
math motivation is depicted in Figure 1.
As can be seen by examining the model in Figure 1, each
observed variable Yi (i.e., with Y1 = M9, Y2 = M10, Y3 = M13,
Y4 = M16, and Y5 = M17) loads on two specific factors, F1
and F2 (called either the Level and Shape factors, or the Intercept and Slope factors, respectively) we note that for Y1 =
M9 the loading on the second factor is set to 0 (indicated by
the dotted line, see further discussion below).

M16

F1

Figure 1.

M17

F2

Unconditional longitudinal model for math motivation.

Based on the theoretical questions considered in this study,


our main interest is in examining childrens initial intrinsic
math motivation (and later math achievement) at age 9 and its
relationship with the rate of change from ages 9 to 17. Consequently, different approaches to the coding of time can be
utilized within the LS and the IS modeling strategy. In particular, with the IS model in order to preserve years as the unit of
time, the component of time is coded in increments of years
(after placing the origin of time at age 9, which entails coding
this age as 0, i.e., age 9). Thus, each coding of age progresses
in increments of the age of assessment. Such a coding scheme
results in the following factor loading matrix for all individuals
for the five assessment occasions and assumes that the trajectory is constant over time (i.e., is linear):
1

1
= 1

1
1

(1)

where yi is the initial status measured at time 1 (also referred


to as the Intercept or Level) of an individuals change trajectory, and yi is the Slope or the Shape (the change in Yi
between the consecutive measurements) of the change trajectory, t corresponds to the measured time points, and it to the
model residual for each individual. Because yi and yi are
random variables, these model parameters can be represented
by a group mean intercept (y) and mean slope (y) plus the
component of individual intercept variation (yi) and slope
variation (yi), as indicated by the following so-called level 2
or between-person model equations for which, as with the
above-mentioned parameters, sample based estimates are
generally obtained:

M13

1
4

7
8

(4)

We note that by fixing the loadings of the five assessment


occasions on the first factor to a value of 1 ensures that it is
interpreted as an initial true (i.e., error-free) status, that is, as
a baseline point of the underlying developmental process
under investigation (as indicated below, this same initial
loading pattern is also used on the first factor in the LS model).
In cases where a quadratic trajectory is assumed, implying
change in the rate of change over time, the following factor
loading matrix would be used:
1

1
= 1

1
1

0 0

1 1
4 16

7 49
8 64

(5)

To represent the quadratic trajectory of data, the factor


loadings for the added third factor (the quadratic factor) are
simply squared loadings of the second factor (the linear
factor). For any assumed higher order polynomial trajectories,
additionally patterned factor loading matrices would be
required.
In contrast to the above-illustrated coding of time, for the
LS modeling strategy the component of time is coded by fixing
the loadings on the second factor as follows (where * corresponds to a freely estimated loading):

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2007, 31 (4), 317327

1
= 1

1
1

*
*

*
1

(6)

Fixing the loading of the last assessment occasion (i.e., Y5 =


M17) on the second factor to a value of 1 and that of the first
assessment occasion on it to a value of 0, ensures that this
factor is interpreted as a change factor (i.e., reflecting the shape
of the change process studied, regardless of the actual trajectory encountered linear, quadratic, cubic, etc.). Freeing the
loadings of the remaining assessment occasions on this factor
implies that they denote the part of overall change that occurs
between the first and each of these later measurement
occasions (Duncan et al., 1999; McArdle & Anderson, 1990).
It is important to note that the loadings on the second factor
when constrained to specific values (as in the case of the IS
model) determine the form of the function (linear, quadratic,
etc.). Consequently, the actual choice of these parameters will
affect the magnitude of the correlation between the two
considered factors, and any interpretation of this correlation
should therefore be made conditionally on the choice of these
time codes. By specifying the change trajectory in increments
of years, as was done in the IS model, the correlation between
the Intercept and Slope factors reflects the relationship
between the initial point (set at age 9) and the slope of the
specifically a priori proposed linear trajectory, which we also
note may or may not be the appropriate model to describe the
change process being examined. Indeed, if such a proposed
linear model does not fit the data well, one generally tries to
use a higher order polynomial trajectory model to fit to the data
(e.g., quadratic, cubic, or higher depending on the available
number of assessments). Unfortunately, such models are
generally extremely difficult to interpret and cannot always
adequately reflect the trajectories actually encountered in the
data. In terms of interpretation, if one considers that the linear
trajectory is modeling constant change over time, the quadratic
trajectory models change in the rate of change over time, the
cubic trajectory models change in the change in the rate of
change over time, and so on, one can quickly see how difficult
it becomes to reach meaningful substantive conclusions
(Bollen & Curran, 2006).
In contrast to the above-stated concerns with the IS
modeling strategy, the LS model is quite easy to interpret (and
is expected to regularly fit the data better). By setting the
parameter at time one to be equal to 0, the parameter of the
final time to be equal to 1 and to freely estimate all the loadings
between the first and final time points, the freed loadings now
reflect the cumulative proportion of total change between two
time points relative to the total change occurring from the first
to the last time point (regardless of the trajectory shape, even
if it is nonlinear), and the correlation between the Level and
Shape factors simply reflects their degree of overlap (Raykov
& Marcoulides, 2006).
When time-varying covariates are considered both as a
separate model and incorporated into a latent change model in
which they can also be studied, the model is more commonly
referred to as a multivariate model (the term multivariate latent
curve model is also used to define such a model, see Bollen &

321

Curran, 2006; Duncan et al., 1999; Raykov & Marcoulides,


2006). Parameterization of such a multivariate model involves
not only the declaration of expressions for the IS or LS factors
of the repeated observations on measures of the covariate (in
this study the math achievement scores, which for ease of
discussion will be symbolized by X), but also a combination of
the separate model concerning the measures of the other
considered variable (in this study the intrinsic math motivation
scores already denoted as Y).
Assuming now that the series of nine repeated measurements on the covariate math achievement are denoted as Xit
(where the index i once again represents each observed individual in the study and t represents the time-ordered measurements on math achievement), the model to describe an
individuals development over the repeated measurements can
be written as:
Xit = xi + xit + it

(7)

where xi is the intercept or level of an individuals change


trajectory, xi is the slope or shape of the change trajectory, t
corresponds to the measured time points, and it to the model
residual for each individual. Subsequently, the group mean
intercept (x) and mean slope (x) along with the component
of individual intercept variation (xi) and slope variation (xi)
can be written as before:
xi = x + xi

(8)

xi = x + xi

(9)

The unconditional latent change model for the nine repeated


measurements on math achievement is depicted in Figure 2
(with the usual assumption of no covariance structure between
the residuals of the longitudinally observed variables). As can
be seen by examining the model in Figure 2, each observed
variable Xi (i.e., with X1 = Ach9, X2 = Ach10, X3 = Ach11,
X4 = Ach12, X5 = Ach13, X6 = Ach14, X7 = Ach15, X8 =
Ach16, X9 = Ach17) loads on the two factors (except that for
X1 = Ach9 the loading on the second factor is set to 0 indicated by the dotted line). Once again, based on the theoretical
questions considered in this study, our interest is in examining
childrens initial math achievement at age 9 and its relationship
with the rate of change from ages 9 to 17. As indicated previously, the coding of time can be approached from the IS or LS
modeling strategies. With the IS model, the component of time
would be coded in increments of years (after placing the origin
of time at age 9, which again entails coding this age as 0). Such
a coding scheme, if it assumes that the trajectory is constant
over time (i.e., is linear) or changes in the rate of change over
time (i.e., is quadratic), would have the following factor
loading matrices, respectively:

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

1
1

1
= 1

1
1

1
1

1
2

3
4

5
6

7
8

(10)

322

GOTTFRIED ET AL. / LATENT CHANGE ANALYSIS OF DECLINE IN MOTIVATION

1
1

1
= 1

1
1

1
1

0
1
2
3
4
5
6
7
8

1
4

9
16

25
36

49
64

(11)

For the LS model, the following factor loading structure would


be used (where * is a freely estimated loading):
1

1
1

1
= 1

1
1

1
1

*
*

*
*

*
*

*
1

(12)

With the LS modeling strategy, fixing the loading of the last


assessment occasion (i.e., X9 = Ach17) on the second factor
to a value of 1 and that of the first assessment occasion to a
value of 0, as discussed earlier, ensures that this factor is interpreted as a change factor (i.e., reflecting the shape of the
change process studied).
Finally, the relationships between the two repeatedly
measured intrinsic math motivation and math achievement
scores are related to one another via the factors of each
measured variable and are based on the covariance structure
of the level 2 models. Such a complete multivariate model
including the covariance structure of the level 2 models is
depicted in Figure 3. This covariance structure can be defined
by the following symmetric matrix :

Ach9

Ach10

y ,y

= y ,y
x ,y

x ,y

Ach11

Ach12

F1

y ,y
x ,y

x ,x

x ,y

x ,x

(13)

where ay,ay = the Intercept or Slope variance for intrinsic math


motivation (what was referred to in Equation 3 as yi), by,by =
the Slope or Shape variance for intrinsic math motivation
(what was referred to in Equation 3 as yi), ax,ax = the Intercept or Level variance of math achievement, xx = the Slope
or Shape variance of math achievement, y,ay = the covariance between the two factors for intrinsic math motivation,
x,ax = the covariance between the two factors for math
achievement, and the remaining values correspond to the
covariances between the two factors for both intrinsic math
motivation and math achievement interchangeably. For ease of
interpretation these covariances can also be computed and
interpreted as correlation coefficients. Consequently, based
upon this multivariate model, the relationship between the
variables X and Y (i.e., math achievement and intrinsic math
motivation) over time is actually determined through the
relationship between the factors of each.

Results
In the following section, we present results from fitting the
proposed IS and LS multivariate models to the data using
LISREL 8.7 (Jreskog & Srbom, 1993) based on full information maximum likelihood (FIML) parameter estimation to
handle the presence of any missing data (Arbuckle, 1996;
although identical results should be obtained using any
currently available SEM programs; e.g., EQS, Mplus, Mx).
The descriptive statistics for the total sample on the observed
variables used in the study are presented in Table 1.
Because we posited an a priori defined model to be tested,
our key interest is initially in the assessment of model fit. Once

Ach13

Ach14

Ach15

F2

Figure 2.

x ,x

Unconditional longitudinal model for math achievement.

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

Ach16

Ach17

323

INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2007, 31 (4), 317327

M9

A9

M10

A10

M13

Mot
F1

Mot
F2

Ach
F1

Ach
F2

A11
Figure 3.

A12

Table 1
Descriptive statistics for intrinsic math motivation and math
achievement
(N = 114)

Intrinsic math motivation


Y1 = M9
Y2 = M10
Y3 = M13
Y4 = M16
Y5 = M17
Math achievement
X1 = A9
X2 = A10
X3 = A11
X4 = A12
X5 = A13
X6 = A14
X7 = A15
X8 = A16
X9 = A17

A13

A14

A15

M17

A16

A17

Multivariate model of math motivation and achievement.

the overall model fit is determined, then the importance and


interpretation of the specific parameter estimates can be more
clearly evaluated. Without a good overall model fit, we might
even have to reconceptualize the proposed theoretical multivariate model. To evaluate model fit, a number of indices were

Variables

M16

SD

100.22
96.72
93.38
84.85
85.43

16.49
16.48
15.50
16.19
16.56

64.02
66.38
88.29
84.40
78.13
74.42
73.18
70.81
64.83

27.58
26.73
14.64
18.45
22.52
24.05
25.17
27.15
27.85

examined. These include the overall 2 goodness-of-fit test, the


comparative fit index (CFI), Akaikes information criterion
(AIC), and the root mean square error of approximation
(RMSEA) along with its associated confidence intervals. These
indices were selected because of their widespread use
(Marcoulides & Hershberger, 1997). Detailed criteria for
evaluation of model fit based on these fit indices can be found
in Byrne (1998), Hu and Bentler (1999), Marcoulides and
Hershberger (1997), and Raykov and Marcoulides (2006). In
summary, it is generally recognized that to support model fit a
consensus among the following is needed: nonsignificant 2
goodness-of-fit value (although because the 2 measure is wellknown to be sensitive to sample size issues and has a tendency
to reject models that are even only marginally inconsistent with
the data, much more emphasis is placed on the other fit
criteria); a CFI > .90; an AIC value closer to the saturated
model than the independence model; an RMSEA below .05
and the left endpoint of its 90% confidence interval is
markedly smaller than .05 (with this interval not excessively
wide; Raykov & Marcoulides, 2006).
The examined proposed linear IS model using the factor
loading matrices in Equations 4 and 10 provided the following
fit criteria: 2(91) = 466.36, p < .05; model AIC = 601.27
(saturated AIC = 602.30, independence AIC = 2891.53);
CFI = .86; and RMSEA = .19 (.17; .20). The proposed LS
model using Equations 6 and 12 provided the following fit
criteria: 2(81) = 173.49, p < .05; model AIC = 437.29 (saturated AIC = 602.30, independence AIC = 2891.53); CFI =
.96; and RMSEA = .06 (.02; .12). An examination of the fit
criteria based upon the above-outlined guidelines indicates
that the proposed linear IS model did not fit the data well as

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

324

GOTTFRIED ET AL. / LATENT CHANGE ANALYSIS OF DECLINE IN MOTIVATION

opposed to the LS model, which fit the data well.1 Based on


the above poor-fitting results for the linear IS model, we also
examined a proposed quadratic IS model. The following fit
criteria were observed for the proposed quadratic IS model:
2(78) = 281.40, p < .05; model AIC = 539.89 (saturated
AIC = 602.30, independence AIC = 2891.53); CFI = .88; and
RMSEA = .16 (.14; .18). Although some degree of model
improvement in the above fit criteria occurred, it was evident
that even this higher order polynomial was not a good fit to the
data. Because the LS model was determined to fit the data
reasonably well and the relative ease with which the parameters
can be interpreted to address the substantive focus of the
study, we decided to only thoroughly assess the significance of
the empirical validation of our proposed LS model. We emphasize again that such poor-fitting IS models are often encountered in empirical applications of multivariate longitudinal
models whenever prespecified trajectories do not always adhere
to an observed developmental trend.
Table 2 presents the LISREL factor loading parameter estimates of the proposed LS model tested. These coefficients
summarize a number of relevant findings with respect to
intrinsic math motivation and math achievement scores. With
respect to intrinsic math motivation, the estimated coefficients
of the Shape factor loadings representing the proportion of
change relative to the total change occurring over all time
points (i.e., 0, .19, .43, 1.04, 1) reflect a decline in intrinsic
math motivation. The decline of the intrinsic math motivation
scores is consistently approximately 20% each year and
appears to start leveling off at age 16 (although we are
assuming that a somewhat similar trajectory pattern occurred
over the years that no assessments were made). Significant
variance existed in both the Level (s2 = 63.53, t = 2.64, p < .05)
and the Shape factors (s2 = 87.00, t = 2.57, p < .05) reflecting
meaningful individual variability in the average initial and
change scores of individuals on intrinsic math motivation over
time. The mean value of the Shape factor was negative and
significantly different from 0 (y = 14.31, t = 8.24, p < .05
which we note here and throughout the results section are
sample based estimated), indicating that there has been over
time a steady decline in intrinsic math motivation scores from
the initial measured and also significantly different from zero
mean value on the Level factor (y = 99.68, t = 67.84,
p < .05). The nonsignificant covariance observed between the
Level and Shape factors (cov = 27.59, t = 1.29, p > .05) shows
that initial levels of intrinsic math motivation scores are not
related to the decreasing levels of motivation scores over time,
and vice versa (see Figure 4, which presents the results for the
covariance structure of the level 2 models).
There is also evidence of significant change in math achievement scores from age 9 to age 17. The estimated coefficients
of the Shape factor loadings representing the proportion of
change over time (i.e., 0, .16, 3.61, 2.62, 1.35, .57, .34,
.21, 1) also reflect nonlinearity in math achievement scores.
The mean value of the Shape factor was negative and significantly different from 0 (x = 5.11, t = 2.73, p < .05),

1 We note that the separate unconditional linear IS model for math intrinsic
motivation did fit the data well, 2(10) = 15.13, p = .13; model AIC = 72.57
(saturated AIC = 86.04, independence AIC = 259.27); CFI = .98; and
RMSEA = .06 (.0; .13), but the unconditional linear IS model for math achievement did not fit the data, 2(48) = 651.55, p < .05; model AIC = 1015.61 (saturated AIC = 258.13, independence AIC = 2060.57); CFI = .69; and RMSEA =
.41 (.38; .44).

Table 2
Factor loading parameter estimates, standard errors, and critical
t ratios for LS model
Factor loadings
Intrinsic math motivation
Time 1 M9
Time 2 M10
Time 3 M13
Time 4 M16
Time 5 M17
Math achievement
Time 1 A9
Time 2 A10
Time 3 A11
Time 4 A12
Time 5 A13
Time 6 A14
Time 7 A15
Time 8 A16
Time 9 A17

Estimate

SE

Critical ratio

0=
0.19
0.43*
1.04*
1=

0.10
0.09
0.07

1.91
4.78
15.16

0=
0.16
3.61*
2.62*
1.35
0.57
0.34
0.21
1=

0.48
1.60
1.25
0.81
0.54
0.46
0.29

0.33
2.26
2.10
1.67
1.05
0.74
0.73

*p < .05.

27.59

Motivation
level

Motivation
shape
85.36*

12.56
101.54*

21.37*

94.00*
Achievement
level

Achievement
shape

*p < .05.
Figure 4. Model of the covariance structure of math motivation and
achievement.

indicating that overall there has been a decline in math achievement scores from the initial measured mean value on the Level
factor (x = 70.72, t = 25.21, p < .05). This also suggests
significant average math achievement decline across the
duration of this study. The significant Level and nonsignificant
Shape variances (s2 = 566.17, t = 5.75, p < .05; and s2 = 21.00,
t = 1.38, p > .05, respectively) indicate marked individual
differences in starting position on math achievement scores
over the course of the study but the degree of individual variability apparently becomes much smaller in the later years.
Finally, the statistically significant covariance observed
between the Level and Shape factors of math achievement
(cov = 94.00, t = 3.06, p < .05; r = .86) suggests that initial
levels of math achievement are related to the changing math
achievement scores over time, and vice versa.
It is also important to note that there is a statistically significant and positive correlation (cov = 101.54, t = 3.25, p < .05;
r = .54) between the Level factors of intrinsic math motivation
and math achievement, as depicted in Figure 4. This shows

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2007, 31 (4), 317327

that the initial average scores on intrinsic math motivation are


related to initial average scores on math achievement. Similarly, there is a statistically significant and positive correlation
between the Shape factors of intrinsic math motivation and
math achievement (cov = 21.37, t = 2.00, p < .05; r = .50).
This shows that the later average scores of participants on
intrinsic math motivation are related to the later average scores
on math achievement. A significant and positive correlation
between the math achievement Level factor and the intrinsic
math motivation Shape factor was found (cov = 85.36, t =
2.38, p < .05; r = .38). This shows that later average scores of
participants on intrinsic math motivation scores are related to
the initial average scores on math achievement. Finally, a
nonsignificant covariance between the intrinsic math motivation Level factor and the math achievement Shape factor
(cov = 12.56, t = 1.61, p > .05) was found.

Discussion
The potential problems of various coding choices pointed out
by previous researchers were considered in this article for the
LS and the IS models. The multivariate LS model was determined to fit the data well, whereas the IS had poor fit. The
results of this investigation based on the LS model supported
the prediction that math achievement is a significant contributor to the developmental decline in intrinsic math motivation.
The data indicated significant changes in both math motivation and achievement from ages 9 to 17 established by models
1 and 2, respectively. These findings are consistent with
previous research as reviewed above. Moreover, the present
research extends beyond the reporting of mean declines in
math motivation and achievement, which, as noted earlier, is
the current state of the developmental literature with regard to
motivational decline (Wigfield et al., 2006). The present
findings indicate that these dual declines are in correspondence
with each other across these years and that math achievement
is a significant contributor to the developmental decrease in
intrinsic math motivation from childhood through adolescence. The results supported the prediction that math achievement significantly contributes to the decline in math
motivation over this age range. This is the first documentation
of the correspondence between developmental declines in
intrinsic math motivation and achievement.
These findings elucidate the contributory role of achievement to motivational decline in the math domain. By studying
achievement directly, we have used an actual index of child
competency. This has been absent in studies of developmental
changes in intrinsic motivational variables. Moreover, the
present data are not confounded by specific school, teacher,
curriculum, or geographic factors because the participants
attended many different schools across the USA rather than a
single school. Had the latter been the case, the developmental
changes in both motivation and achievement could have been
related to school specific factors.
Regarding the age range studied, our sample spans a broad
range from elementary school through the end of high school.
Therefore, the declines obtained herein are long-term, as is the
contribution of achievement to motivational decline. That is,
by studying these decreases across this age span, we can see
that as early as elementary school, achievement plays a significant role in the ensuing decline in both math intrinsic motivation and achievement. Therefore, this deterioration begins

325

early in the school life of children. This is alarming because of


the current critical need to create math competence in US
students.
Examination of the significant paths in Figure 4 suggests the
following developmental course of motivation and achievement over time. Concurrent initial levels of motivation and
achievement bear a significant relationship to each other, as do
the ending points. Given the previously discussed nonsignificant covariance observed between the Level and Shape factors
of math motivation (and the determination that initial levels of
math motivation are not related to the decreasing levels of
motivation scores over time, and vice versa), the positive correlation between both the math motivation and math achievement Level and Shape factors indicates that initial math
motivation and achievement are related to later changes in
math achievement and motivation, but in the case of motivation the relationship is indirect (through achievement),
whereas in the case of achievement it is direct. Therefore,
regarding motivation, ones beginning level is related to ones
ending level, but only indirectly through its relationship to
achievement. A significant direct path was obtained from the
beginning level of math achievement to the decline, or shape,
of motivation through age 17. This path augments the finding
of a direct path from achievement level to achievement shape
(change) by showing its pervasiveness with regard to motivational change as well. Therefore, childrens level of math
achievement as early as age 9 is a significant factor for their
ultimate level of achievement as well as motivation through the
end of high school. Whereas motivation plays a significant,
albeit indirect, role in developmental changes in both achievement and motivation, achievement itself is directly related to
both domains. This finding implies that childrens early level
of math success has long-term consequences for their future
math achievement, as well as intrinsic motivation
The present findings imply that poorer initial levels of math
achievement place students at-risk for long-term declines in
both math achievement and motivation. This finding augments
our previous assertion that those who enter adolescence with
lower academic intrinsic motivation are in jeopardy for longterm motivational adversity (A.E. Gottfried et al., 2001). The
present findings complement this concept of developmental
jeopardy, and add to it by identifying the early level of achievement as a significant and contributory factor.
These findings also support the theoretical relationships
expected to exist between intrinsic motivation and achievement (Berlyne, 1971; A.E. Gottfried, 1985; Hunt, 1971). Thus
far, relationships between academic intrinsic motivation and
achievement, as well as general motivation and achievement,
have been found to be positive, but have not involved LCA
models. This investigation goes beyond by establishing the
longitudinal correspondence between the developmental
changes in math achievement and intrinsic motivation
domains.
The results revealed that the sample was less variable, or
more homogeneous, at the end compared with the beginning
points in both motivation and achievement. This is interesting
developmentally and educationally. Developmentally, this
decrease in variability would be expected if one considers that
as children have increasing experience with their academic
strengths and weaknesses over the years, they would be
expected to focus in on particular areas. Therefore, their variability in both motivation and achievement would be expected
to decrease over time as they focus more on particular

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

326

GOTTFRIED ET AL. / LATENT CHANGE ANALYSIS OF DECLINE IN MOTIVATION

domains. Educationally, this may direct students choices of


courses in high school, college majors, and career paths.
Finally, we anticipate that this research will inform the
development of appropriate educational interventions to both
reverse and prevent these developmental declines. This is
essential for the academic and personal well-being of students,
for adequate preparation of students to enter higher education,
as well as for the integrity of the development of math and
related science professionals in the USA.

References
Anderman, E.M., & Maehr, M.L. (1994). Motivation and schooling in the
middle grades. Review of Educational Research, 64, 287309.
Arbuckle, J.L. (1996). Full information estimation in the presence of incomplete
data. In G.A. Marcoulides & R.E. Schumacker (Eds.), Advanced structural
equation modeling: Issues and techniques (pp. 243277). Mahwah, NJ: Erlbaum.
Berlyne, D.E. (1971). What next? Concluding summary. In H.I. Day,
D.E. Berlyne, & D.E. Hunt (Eds.), Intrinsic motivation: A new direction in
education (pp. 186196). Toronto: Holt, Rinehart & Winston.
Biesanz, J.C., Deeb-Sossa, N., Papadakis, A.A., Bollen, K.A., & Curran, P.J.
(2004). The role of coding time in estimating and interpreting growth curve
models. Psychological Methods, 9, 3052.
Bollen, K.A., & Curran, P.J. (2006). Latent curve models: A structural equation
approach. New York: Wiley.
Byrne, B.M. (1998). Structural equation modeling with LISREL, PRELIS, and
SIMPLIS: Basic concepts, applications, and programming. Mahwah, NJ:
Erlbaum.
Ding, C.S., & Davison, M.L. (2004). A longitudinal study of math achievement
gains for initially low achieving students. Contemporary Educational Psychology,
30, 8195.
Duncan, T.E., Duncan, S.C., Strycker, L.A., Li, F., & Alport, A. (1999). An
introduction to latent variable growth curve modeling. Mahwah, NJ: Erlbaum.
Eccles, J.S., & Midgley, C. (1989). Stage-environment fit: Developmentally
appropriate classrooms for young adolescents. In C. Ames & R. Ames (Eds.),
Research on motivation in education:Vol. 3, Goals and cognitions (pp. 139186).
New York: Academic Press.
Eccles, J.S., & Midgley, C. (1990). Changes in academic motivation and selfperception during early adolescence. In R. Montemayor, G.R. Adams, & T.P.
Gulotta (Eds.), From childhood to adolescence: A transitional period?
(pp. 134155). Newbury Park, CA: Sage.
Eccles, J.S., Wigfield, S., & Schiefele, U. (1998). Motivation to succeed. In
N. Eisenberg (Vol. Ed.), Handbook of child psychology:Vol. 3, Social, emotional
and personality development (5th ed., pp. 10171095). New York: Wiley.
Elliott, A.J., & Dweck, C.S. (2005). Handbook of competence and motivation. New
York: Guilford Press.
Fredericks, J.A., & Eccles, J.S. (2002). Childrens competence and value beliefs
from childhood through adolescence: Growth trajectories in two male-sextyped domains. Developmental Psychology, 38, 519533.
Gottfried, A.E. (1985). Academic intrinsic motivation in elementary and junior
high school children. Journal of Educational Psychology, 77, 631645.
Gottfried, A.E. (1986). Childrens Academic Intrinsic Motivation Inventory. Lutz,
FL: Psychological Assessment Resources.
Gottfried, A.E. (1990). Academic intrinsic motivation in young elementary
school children. Journal of Educational Psychology, 82, 525538.
Gottfried, A.E., Fleming, J.S., & Gottfried, A.W. (1994). Role of parental motivational practices in childrens academic intrinsic motivation and achievement. Journal of Educational Psychology, 86, 104113.
Gottfried, A.E., Fleming, J.S., & Gottfried, A.W. (2001). Continuity of academic
intrinsic motivation from childhood through late adolescence: A longitudinal
study. Journal of Educational Psychology, 93, 313.
Gottfried, A.E., & Gottfried, A.W. (2004). Toward the development of a conceptualization of gifted motivation. Gifted Child Quarterly, 48, 121132.
Gottfried, A.W. (1985). Measures of socioeconomic status in child development
research: Data and recommendations. Merrill-Palmer Quarterly, 31, 8592.
Gottfried, A.W., & Gottfried, A.E. (1984). Home environment and cognitive
development in young children of middle-socioeconomic-status. In A.W.
Gottfried (Ed.), Home environment and early cognitive development: Longitudinal research (pp. 57115). New York: Academic Press.
Gottfried, A.W., Gottfried, A.E., Bathurst, K., & Guerin, D. (1994). Gifted IQ:
Early developmental aspects. New York: Plenum Press.
Gottfried, A.W., Gottfried, A.E., Bathurst, K., Guerin, D.W., & Parramore,

M.M. (2003). Socioeconomic status in childrens development and family


environment: Infancy through adolescence. In M.H. Bornstein & R.H.
Bradley (Eds.), Socioeconomic status, parenting, and child development
(pp. 189207). Mahwah, NJ: Erlbaum.
Gottfried, A.W., Gottfried, A.E., Cook, C., & Morris, P. (2005). Educational
characteristics of adolescents with gifted academic intrinsic motivation: A
longitudinal study from school entry through early adulthood. Gifted Child
Quarterly, 49, 172186.
Gottfried, A.W., Gottfried, A.E., & Guerin, D.W. (2006). The Fullerton Longitudinal Study: A long-term investigation of intellectual and motivational giftedness. Journal for the Education of the Gifted, 29, 430450.
Guerin, D.W., Gottfried, A.E., Oliver, P.H., & Thomas, C.W. (2003). Temperament: Infancy through adolescence. New York: Kluwer Academic/Plenum Press.
Guerin, D.W., & Gottfried, A.W. (1994). Developmental stability and change in
parent reports of temperament: A ten-year longitudinal investigation from
infancy through preadolescence. Merrill-Palmer Quarterly, 40, 413421.
Harter, S. (1981). A new self-report scale of intrinsic versus extrinsic motivation in the classroom: Motivational and informational components. Developmental Psychology, 17, 300312.
Hollingshead, A.B. (1975). Four factor index of social status. Unpublished manuscript, Department of Sociology, Yale University, New Haven, CT.
Hu, L.-T., & Bentler, P.M. (1999). Cutoff criteria for fit indices in covariance
structure analysis: Conventional criteria versus new alternatives. Structural
Equation Modeling, 6, 155.
Hunt, J. McV. (1971). Intrinsic motivation and psychological development. In
H.M. Schroder & P. Suedfeld (Eds.), Personality theory and information processing (pp. 131177). New York: Ronald Press.
Jacobs, J.E., Lanza, S., Osgood, D.W., Eccles, J.S., & Wigfield, A. (2002).
Changes in childrens self-competence and values: Gender and domain differences across grades one through twelve. Child Development, 73, 509527.
Jreskog, K.G., & Srbom, D. (1993). LISREL8: Users reference guide. Chicago,
IL: Scientific Software.
Kelly, D.L. (2002). The TIMSS 1995 international benchmarks of mathematics
and science achievement: Profiles of world class performance at fourth and
eighth grades. Educational Research and Evaluation, 8, 4154.
Lepper, M.R., Corpus, J.H., & Iyengar, S.S. (2005). Intrinsic and extrinsic motivational orientations in the classroom: Age differences and academic correlates. Journal of Educational Psychology, 97, 184196.
Lepper, M.R., & Henderlong, J. (2000). Turning play into work and work
into play: 25 years of research on intrinsic versus extrinsic motivation. In
C. Sansone & J.M. Harackiewicz (Eds.), Intrinsic and extrinsic motivation:
The search for optimal motivation and performance (pp. 257307). New York:
Academic Press.
Lepper, M.R., Sethi, S., Dialdin, D., & Drake, M. (1997). Intrinsic and extrinsic motivation: A developmental perspective. In S.S. Luthar, J.A. Burack,
D. Cicchetti, & J.R. Weisz (Eds.), Developmental psychopathology: Perspectives
on adjustment, risk, and disorder (pp. 2350). New York: Cambridge University
Press.
Marcoulides, G.A., & Hershberger, S.L. (1997). Multivariate statistical methods:
A first course. Mahwah, NJ: Erlbaum.
McArdle, J.J. (1988). Dynamic but structural equation modeling of repeated
measures data. In J.R. Nesselroade & R.B. Cattell (Eds.), Handbook of multivariate experimental psychology (2nd ed., pp. 561614). New York: Plenum
Press.
McArdle, J.J., & Anderson, E. (1990). Latent growth models for research on
aging. In J.E. Birren & K.W. Schaie (Eds.), Handbook of the psychology of aging
(3rd ed., pp. 21144). New York: Academic Press.
Middleton, J.A., & Spanias, P.A. (1999). Motivation for achievement in mathematics: Findings, generalizations, and criticisms of the research. Journal for
Research in Mathematics Education, 30, 6588.
National Academy of Sciences (2005). Rising above the gathering storm: Energizing and employing America for a brighter economic future. Available [accessed 13
April 2006]: http://www.nap.edu/catalog/11463.html
ODwyer, L.M. (2005). Examining the variability of mathematics performance
and its correlate using data from TIMSS 95 and TIMSS 99. Educational
Research and Evaluation, 11, 155177.
Raykov, T., & Marcoulides, G.A. (2006). A first course in structural equation
modeling (2nd ed.). Mahwah, NJ: Erlbaum.
Rogosa, D.R., & Willett, J.B. (1985). Understanding correlates of change by
modeling individual differences in growth. Psychometrika, 50, 203228.
Shen, C. (2002). Revisiting the relationship between students achievement and
their self-perceptions: A cross-national analysis based on TIMSS 1999 data.
Assessment in Education, 9, 161184.
Spinath, B., & Spinath, F.M. (2005). Longitudinal analysis of the link between
learning motivation and competence beliefs among elementary school
children. Learning and Instruction, 15, 87102.

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

INTERNATIONAL JOURNAL OF BEHAVIORAL DEVELOPMENT, 2007, 31 (4), 317327


Tatsuoka, K.K., Corter, J.E., & Tatsuoka, C. (2004). Patterns of diagnosed
mathematical content and process skills in TIMSS-R across a sample of 20
countries. American Educational Research Journal, 41, 901926.
Uguroglu, M.E., & Walberg, H.J. (1979). Motivation and achievement: A quantitative synthesis. American Educational Research Journal, 16, 375389.
White, R.W. (1959). Motivation reconsidered: The concept of competence.
Psychological Review, 66, 297333.
Wigfield, A., Eccles, J.S., Schiefele, U., Roeser, R.W., & Davis-Kean, P. (2006).
Development of achievement motivation. In N. Eisenberg (Ed.), Handbook of

327

child psychology:Vol. 3. Social, emotional, and personality development (6th ed.,


pp. 9331002). New York: Wiley.
Wigfield, A., & Wagner, A.L. (2005). Competence, motivation, and identity
development during adolescence. In A.J. Elliot & C.S. Dweck (Eds.), Handbook
of competence and motivation (pp. 222239). New York: Guilford Press.
Woodcock, R.W., & Johnson, M.B. (1977). WoodcockJohnson Psycho-Educational
Battery. Hingham, MA: Teaching Resources.
Woodcock, R.W., & Johnson, M.B. (1989). WoodcockJohnson Psycho-Educational
Battery Revised. Allen, TX: DLM.

Downloaded from jbd.sagepub.com at ANDERSON UNIV on October 6, 2014

You might also like